implementation of a process-based catchment model in a ... · oped for the himalayan conditions. it...

Hydrol. Earth Syst. Sci., 11, 1323–1339, 2007www.hydrol-earth-syst-sci.net/11/1323/2007/© Author(s) 2007. This work is licensedunder a Creative Commons License.

Hydrology andEarth System

Sciences

Implementation of a process-based catchment model in a poorlygauged, highly glacierized Himalayan headwater

M. Konz1, S. Uhlenbrook2, L. Braun3, A. Shrestha4, and S. Demuth5,6

1University of Basel, Department of Environmental Sciences, Applied and Environmental Geology, Basel, Switzerland2UNESCO-IHE, Department of Water Engineering, Delft, The Netherlands3Bavarian Academy of Sciences, Commission of Glaciology, Munich, Germany4Department of Hydrology and Meteorology, Snow and Glacier Hydrology Unit, Katmandu, Nepal5University of Freiburg, Institute of Hydrology, Freiburg, Germany6IHP/HWRP Secretariat, Federal Institute of Hydrology, Koblenz, Germany

Received: 31 October 2006 – Published in Hydrol. Earth Syst. Sci. Discuss.: 13 November 2006Revised: 5 April 2007 – Accepted: 25 April 2007 – Published: 15 May 2007

Abstract. The paper presents a catchment modeling ap-proach for remote glacierized Himalayan catchments. Thedistributed catchment model TACD, which is widely basedon the HBV model, was further developed for the applicationin highly glacierized catchments on a daily timestep and ap-plied to the Nepalese Himalayan headwater Langtang Khola(360 km2). Low laying reference stations are taken for tem-perature extrapolation applying a second order polynomialfunction. Probability based statistical methods enable bridg-ing data gaps in daily precipitation time series and the re-distribution of cumulated precipitation sums over the previ-ous days. Snow and ice melt was calculated in a distributedway based on the temperature-index method employing cal-culated daily potential sunshine durations. Different meltingconditions of snow and ice and melting of ice under debrislayers were considered. The spatial delineation of hydro-logical response units was achieved by taking topographicand physiographic information from maps and satellite im-ages into account, and enabled to incorporate process knowl-edge into the model. Simulation results demonstrated thatthe model is able to simulate daily discharge for a periodof 10 years and point glacier mass balances observed in theresearch area with an adequate reliability. The simple but ro-bust data pre-processing and modeling approach enables thedetermination of the components of the water balance of a re-mote, data scarce catchment with a minimum of input data.

Correspondence to:M. Konz([email protected])

1 Introduction

The prediction of runoff from Himalayan headwaters is a cru-cial element for sustainable development of the Himalayancountries (e.g. Tarar, 1982; Kattelmann, 1993; Viviroli et al.,2004). Therefore, in 1987 the German Agency for TechnicalCooperation (GTZ) initiated a project to quantify climato-logical and hydrological conditions of Himalayan headwa-ters in Nepal (Grabs and Pokhrel, 1993). Today, the Depart-ment of Hydrology and Meteorology (DHM) is responsiblefor data collection and maintenance of the stations. Dailyreadings of temperature, precipitation and discharge data areoutsourced to local farmers living close to the stations. DHMis doing emergency field trips and regular trips to ensure thereliability of the measurements. In spite of these efforts datalosses occur due to inaccessibility of the catchment becauseof the weather conditions. However, these data are uniquesources for the assessment of water resources of high Hi-malayan catchments. One way to improve water resourcesassessment is to combine existing field data with adequatemodelling approaches, as this has been done in temperateclimate study areas (e.g. Uhlenbrook and Leibundgut 2002).Simulation of snow and ice melt and its impact on the wa-ter balance of a catchment is of vital interest of the scien-tific community (Verbunt et al., 2003). Sophisticated, dataintensive models have been developed for distributed model-ing of glaciated catchments (e.g. Klok et al., 2001). How-ever, classical approaches to simulate snow and ice melt,like the application of the temperature-index method, are stillused because of its robustness especially in data scarce re-gions (Hock, 2003; Zappa et al., 2003). First simulations ofdaily discharge of the Nepalese Himalayan Langtang Kholacatchment were conducted by Braun et al. (1993) using a

Published by Copernicus Publications on behalf of the European Geosciences Union.

1324 M. Konz et al.: Modelling approach for a poorly gauged Himalayan headwater

Fig. 1. Nepal and the Langtang Khola catchment.

conceptual precipitation-runoff model. In this study the au-thors had to apply a constant daily value to simulate the rela-tively high winter discharge. The value was found by calibra-tion and added to the calculated daily runoff value. It turnedout that with the lumped runoff routine of the model it wasnot possible to simulate the annual course of the hydrographwithout this synthetic adjustment.

The catchment model TACD (tracer aided catchmentmodel, distributed) which was originally developed forbasins in the Black Forest Mountains, Germany (Uhlenbrooket al., 2004; Ott and Uhlenbrook, 2004) was further devel-oped for the Himalayan conditions. It is a fully distributed,modular catchment model, which at its core has a process-based runoff generation routine based on dominant processconceptualizations.

Since data availability is limited in remote Himalayanheadwaters suitable ways to regionalize input data and toconceptualize the dominant runoff generation processes haveto be developed. The paper describes the first applicationof this process-based model to a high mountain environmentwith very limited data and an added glacier module. Thespecific objectives are to describe the entire simulation pro-cedure that consists of (i) input data pre-processing to bridgegaps in time series of precipitation and temperature, (ii ) thedelineation of hydrological response units based on topo-graphic and physiographic data, and (iii ) the process-basedrunoff modeling itself. The results of this approach are il-lustrated for the Langtang Khola catchment in Nepal, anddiscussed with respect to process-based catchment model-ing. The methods for input data preparation are describedin detail with the relatively good data set of the LangtangKhola catchment to demonstrate the suitability and applica-

Table 1. Main characteristics of the investigated catchment duringthe investigation period 1987–1998.

Area

Total (km2) 360.0Glacierized (km2/%) 166.1/46.1Debris-covered glacier (km2/%) 32.1/19.31

AltitudesRange (m a.s.l.) 3800–7234Average (m a.s.l.) 5169ExpositionNorth2 (%) 21.3South3 (%) 26.8East, West, Horizontal4 (%) 51.9Mean slope (◦) 26.7Land coverGlacier (km2/%) 166.1/46.1Barren land (km2/%) 183.8/51.1Forest (km2/%) 2.1/0.6Grass land (km2/%) 3.4/0.9Others (km2/%) 4.6/1.3

1 in percent of glacier-covered area2 315◦–45◦3 135◦–225◦4 45◦–135◦ and 225◦–315◦

bility of the methods. The aim was to develop tools that are inprincipal applicable to much less intensively observed basins(poorly gauged or ungauged basins) then the Langtang Kholacatchment.

2 Study site

The 360 km2 high mountain Himalayan Langtang Kholaheadwater is located approx. 100 km north of Katmandu(Fig. 1, Table 1). The catchment reaches from 3800 m a.s.l.up to the Langtang Lirung at 7234 m a.s.l. with two otherpeaks above 7000 m a.s.l.

The average altitude is 5169 m a.s.l., 166 km2 (46%) of the360 km2 catchment area are occupied by glaciers of which32 km2 (19%) are covered by debris, especially the glaciertongues below 5200 m a.s.l. The glacier area is differentto the one published in Braun et al. (1993) where glacierarea was determined to be 38%. They derived the glacierarea from the DAV map (Alpenvereinskartographie, 1990),whereas in this paper the glacier inventory of the Inter-national Center of Integrated Mountain Development (ICI-MOD), Katmandu, Nepal (UNEP/ICIMOD, 2002) was used.

The main valley is dissected by the Langtang KholaRiver and it is typically U-shaped. Table 1 summarizesthe main characteristics of the investigation area. Severalsets of moraines occupy the valley bottom, which have been

Hydrol. Earth Syst. Sci., 11, 1323–1339, 2007 www.hydrol-earth-syst-sci.net/11/1323/2007/

M. Konz et al.: Modelling approach for a poorly gauged Himalayan headwater 1325

attributed to the Little Ice Age, Neoglacial and Late glacial(Heuberger et al., 1984; Ono, 1986). Boulders and screecover the steep slopes and high plateaus, while the occur-rence of forest and grassland in the lower altitudes with lesssteep slopes along the river is limited to 1.5% of the catch-ment area. The riverbed consists of sand and gravel. The areais part of the Main Central Thrust Zone, and its geology con-sists of granite, gneiss and schist. Detailed soil informationis not available for the catchment.

The characteristic feature of the climate of Nepal is themonsoon circulation with predominant easterly winds in thesummer and westerly winds from October to May. In theLangtang Khola catchment, 74% of annual precipitation of615 mm a−1 at the Kyangjing station (3920 m a.s.l., Fig. 1)of the Snow and Glacier Hydrology Unit (SGHU) of the De-partment of Hydrology and Meteorology (DHM) falls dur-ing monsoon season (May till October; mean over periodfrom 1988 to 1998). From June to August total precipita-tion amounts are large and precipitation occurs almost everyday. However, the daily amount generally does not exceed20 mm d−1. In the later part of the monsoon season (Septem-ber to October) the maximum daily amount is considerablyhigher, while the number of rainy days decreases graduallycompared to previous months. In dry season (Novemberto May), precipitation (mainly snow) occurs on only a fewdays, it is produced by the occasional passage of westerlytroughs (“western disturbances”; Ramage, 1971). In general,the precipitation amounts increase with altitude during boththe monsoon and the dry season (Seko, 1987).

Mean daily air temperature is 0.2◦C from October to June,and 8.4◦C during the monsoon season at the meteorologi-cal station (3920 m a.s.l.; Fig. 1). The long-term mean of airtemperature amounts to 1.8 ˚ C for the period 1988 to 1998(Konz et al., 2006a). Shiraiwa et al. (1992) reported meanmonthly air temperatures of approximately−10◦C duringwinter periods at three temporary stations in the valley at al-titudes between 5090 m a.s.l. and 5180 m a.s.l. and differentexpositions during their investigation period from June 1989to March 1991. Air temperature shows distinct altitude de-pendence with a mean annual air temperature from June 1989to March 1991 of 1.9◦C at the SGHU station and−3.5◦C in5090 m a.s.l. A mean annual elevation gradient can be esti-mated as−0.46◦C 100 m−1 (Shiraiwa et al., 1992). Sakai etal. (2004) observed a lapse rate of−0.5◦C 100 m−1 for themonsoon season 1996.

The discharge regime can be classified as glacial withmaximum discharges in July and August and minimum dis-charges during the winter season. Winter discharge is charac-terized by a rather constant base flow with negligible inflowsof rainwater or melt water as the air temperature is generallybelow the melting point. The mean discharge from Decem-ber to the end of April is 2.8 m3 s−1 which constitutes about17% of the annual discharge (Konz et al., 2006a).

Table 2. Summary of DHM reference stations in the Langtang re-gion used for data processing (P : precipitation,T : temperature).

Station Elevation (m a.s.l.) Measurements

Timure 1900 P

Sarmathang 2625 P

Kathmandu Airport 1336 T , P

Thamachit 1847 P

Paigutary unknown P

Dhunche 1982 T , P

Tarke Ghyang 2480 P

3 Data

3.1 Preparation of input data

There is only one meteorological station in the LangtangKhola catchment at 3920 m a.s.l. Air temperature is mea-sured and analyzed in 6 h intervals and published as maxi-mum, minimum and mean daily air temperature. Precipita-tion is read manually and published as daily totals in year-books by the Ministry of Science and Technology, Depart-ment of Hydrology and Meteorology (DHM), Katmandu,Nepal. The data base for this study is the period 1987–1997 as further data are not available or of unreliable qual-ity. Intensive quality analyses of temperature and precipita-tion time series were conducted for each year under inves-tigation. Daily air temperature data are available for 98%and daily sums of precipitation are available for 93% of thedays during the investigation period. Most of the missingvalues occur during the dry season. For some of the datagaps of the precipitation time series sums over several dayswere recorded instead of daily values. Appropriate proce-dures to fill the missing values and to redistribute cumulatedprecipitation sums over the previous days had to be foundbased on records of reference stations of the Standard DataMeteorological Service Network of DHM in the vicinity (upto 60 km away from the catchment) of the Langtang valley.Table 2 summarizes the reference stations with elevation andconducted measurements. The approaches are also applica-ble to catchments with worse data availability then the Lang-tang Khola catchment as shown in Konz et al. (2006). Inthe following we describe pre-processing methods to preparecontinuous time series of precipitation and temperature dataas point information at the meteorological station. The meth-ods are discussed in Sect. 7.1. These time series are the inputdata for the model. Regionalization of these data is realizedin the model itself and therefore discussed in Sect. 4.1.

Air temperature: The filling of missing temperature data(TSGHU) is carried out using a second order polynomialfunction (Weber 1997), which depends only on temperature

www.hydrol-earth-syst-sci.net/11/1323/2007/ Hydrol. Earth Syst. Sci., 11, 1323–1339, 2007


overflow

Input

Runoff

Perculation

Limit of upper storage

Storage capacity

Runoff

Storage capacity

overflow

Input

Runoff

Storage capacity Runoff

Storage capacity

Input

overflow Limit of upper storage

nRGTypes 1 and 2 nRGTypes 3 and 4

Fig. 2. Conceptualization of reservoirs of the different units of runoff generation types with extended river network.

values of the reference stations (TRef):

TRef = AT 2SGHU + BTSGHU + C (1)

The parameters (A,B,C) of the equation are regressively de-termined from mean monthly air temperature values of thereference station and the SGHU station in the catchment.Only those months were considered where no or minimal(less than three values) data gaps occurred. Although the co-efficients were determined for each year individually it canbe stated that the parameters only vary slightly from year toyear. Therefore, average parameters calculated for years withadequate data availability can be used for years with a largenumber of missing data. Data quality and availability of thereference station is highly important, thus we used the Kat-mandu airport station as reference station.

Precipitation: Low lying reference stations (Table 2) areused for filling the missing data. These are the only stationsin the region which provide daily precipitation data. Theprocedure of filling the missing data and redistributing thecumulated values is based on a statistical method, developedby Weber (1997; quoted in Braun et al., 1998). Precipita-tion data of SGHU station and the reference stations wereanalyzed according to the following criteria:

1. the occurrence probability of precipitation events foreach station and for each month;

2. the joint probability of precipitation occurrence at theSGHU station and the DHM reference stations;

3. the mean ratio of monthly sums of precipitation at theSGHU station and the DHM reference stations, and

4. the “true” monthly sums of precipitation at the SGHUstation.

These information are incorporated in a weighting function(Eq. 2) containing the mean ratio of amount of precipitationbetween reference stationn and SGHU station of the monthm (Fn,m) and the weighting of reference stationn accordingto the probability of joint occurrence of precipitation at thetarget station and reference station for the monthm (Wn,m).The last step (iv, in the listing above) often involves estima-tions if missing data occur. The value is used to match thesimulated daily precipitation value to a more realistic value.

PSGHU =1

N·

N∑n=1

(Pn · Fn,m · Wn,m

)(2)

PSGHU: extrapolated precipitation at SGHU station(mm d−1)

Pn: precipitation at reference station (mm d−1)

N: Number of reference stations (–)



Other spatial information: Topographic information wasderived from digital maps of the Survey Department ofNepal and of the topographical map of the Deutscher Alpen-verein (DAV) (Alpenvereinskartographie, 1990). We de-veloped a digital terrain model (DTM) in a resolution of200×200 m2 with the help of a Triangular Irregular Network(TIN) based on the digital contour maps of the Survey De-partment. The vector maps of physiographic features likeglacier area, debris covered area or non-glacier covered areawere converted into raster maps of the same resolution as theDTM. Glacier covered area was taken from maps publishedin the glacier inventory of the International Center of Inte-grated Mountain Development (ICIMOD), Katmandu, Nepal(UNEP/ICIMOD, 2002). The provided digital river networkcovers the glacier free part of the catchment. However, ithas to be assumed that a sub-glacial channel network ex-ists. Field visits have substantiated this assumption. There-fore, the river network was manually extended below glaciers(Fig. 2). This caused a significant improvement of the simu-lation of the dynamics of the hydrograph.

3.2 Hydrological and glaciological data for model calibra-tion and validation

Tracer dilution methods are used to measure discharge as de-scribed by Spreafico and Grabs (1993). SGHU continuouslyperforms these measurements several times per year duringthe dry and rainy seasons. Daily mean discharge (Q) is de-rived from water level readings (8 h) and stage-discharge re-lations. The following Eq. (3) was found to be most suitablefor calculation of discharge values (R2=0.92).

Q = 8.409(h − 0.078)1.334 (3)

Glacier mass balances were observed from 19/5/1996to 6/10/1996 by Fujita et al. (1998) in altitudes from5150 m a.s.l. to 5390 m a.s.l. (Table 3) on debris-free Yalaglacier (Fig. 1) in the Langtang Khola catchment. Braun etal. (1993) published mass balances for the period 28/03/1991to 17/03/1992 of the same glacier.

4 The modeling approach

The modeling approach is based on the model TACD (traceraided catchment model, distributed), which is a process-based catchment model with a modular structure. It can beseen as a modified, fully distributed version of the well-knowHBV-model with a more process-based runoff generationroutine, that uses a spatial delineation of units with the samedominating runoff generation processes. A detailed modeldescription is given by Uhlenbrook et al. (2004) or Uhlen-brook and Sieber (2005). In this study 200×200 m2 grid cellsand daily time steps were used. The model routes the wa-ter laterally by applying the single-flow direction algorithm(O’Callaghan and Mark, 1984). The model is coded within

Table 3. Glacier mass balances in the Langtang Khola catchmentduring March 1991–March 1992 (Braun et al., 1993) and May–October 1996 (Fujita et al., 1998).

Altitude of point Mass Observation periodmeasurement balance

(m a.s.l.) (mm)

5240 −2240 March 1991–March 19925580 510 March 1991–March 19925150 −2300 May–October 19965190 −2100 May–October 19965230 −1700 May–October 19965280 −1500 May–October 19965350 −300 May–October 19965380 300 May–October 19965390 900 May–October 1996

Areal average −357 May–October 1996

the geographical information system PC-Raster (Karssen-berg et al., 2001), which offers a dynamic modeling lan-guage. Since the highly sophisticated original TACD modelcannot be applied to catchments with only temperature andprecipitation data, substantial modifications of the modelwere necessary. The model was further simplified and ideasof the HBV-ETH model presented in Braun et al. (1993) wereimplemented. However, major differences in the spatial dis-cretization and in the simulation of the runoff generation pro-cesses remain in comparison to the HBV-ETH model. Theglacier module with a distributed snow and ice melt calcula-tion was newly introduced into the model.

4.1 Regionalization of meteorological input data

Regionalization of the pre-processed input data is done inthe model and is based on vertical and horizontal gradientstaken from the literature (Sakai et al., 2004; Shiraiwa et al.,1992). According to Shiraiwa et al. (1992) precipitation de-creases from the west to the upper parts of the valley in thenorth-east due to a mountain barrier running west-east at thesouthern side of the valley, which causes rain shadowing ef-fects in the northern part. This is considered in the modelusing a horizontal precipitation gradient of−3.0% 1000 m−1

calculated from the data published by Shiraiwa et al. (1992).In order to compensate for systematic errors in data collec-tion and the lack of representativity of the climatic station,an empirical precipitation correction factorPCF (–) for rainwas introduced. Air temperature is distributed with elevationby applying a fixed lapse rate as given by Sakai et al. (2004)(TGrad(−0.5◦C 100 m−1); see section above). Maximum po-tential evapotranspiration is defined by the parameterETmax(mm d−1) peaking on 1 August.ETmax is the amplitude of asinusoidal function with a minimum of 0 mm d−1 on 1 Febru-ary, thus the seasonal variations of the evapotranspiration



could be simulated. Potential evapotranspiration is region-alized using a fixed, vertical lapse rate of−1.0% 100 m−1

(Sakai et al., 2004).

4.2 Snow and glacier routine

The state of precipitation (liquid or solid) is determinedbased on a temperature thresholdTT (◦C), such that at eachgrid cell the precipitation is modeled as snow if the air tem-perature is belowTT and as rain if the air temperature isaboveTT. A snowfall correction factorSFCF (–), analogueto PCF, is introduced that accounts for systematic measure-ment errors during snowfall. Glaciers are considered as infi-nite water storages. To simulate snow and ice melt we use thewell-known temperature-index method (Zingg, 1951; Braunet al., 1993; Hock, 2003) in a distributed way as shown inEq. (4).

Qmelt = CFMAX × (TSGHU − T T )RexpMapRiceRdebris (4)

Qmelt: Melt water (mm d1)CFMAX: Degree-day factor (mm d−1 ◦C−1)

TSGHU: Daily mean air temperature (◦C)TT: Threshold value of temperature (◦C)RexpMap: Temporal and spatial variable multiplicative factor(–)Rdebris: Parameter for reduced melt water production underdebris cover (–)Rice: Parameter for accelerated melt of ice compared tosnow (–)

Emphasis is placed on modeling the temporal, spatial dis-tribution of melt water production as a function of sunshineduration of a day per grid cell. Consequently, the degree-day factor is temporally and spatially distributed accordingto sunshine duration. This conceptual solution of adaptingHock’s idea (Hock, 1999) allows us to simulate melt in a dis-tributed way without the need of additional observations.

Potential sunshine duration is calculated using the modelPOTRAD (Dam, 2000). This model calculates the potentialsunshine duration on each grid cell for the given time stepof one day, taking topography (slope, aspect, and shadowingeffects) and solar geometry (declination of the sun, latitude,and azimuth angle). A linear relation is used for the transfor-mation of potential sunshine duration into the multiplicativefactor for the temperature-index methodRexpMap(–) (Eq. 5).

RexpMap= ((Rexp − (1/Rexp))/(Shademax−Shademin))

×(Shade−Shademin) + (1/Rexp) (5)

Rexp: Factor for cells with maximum potential sunshine du-ration (–)Shademax: Maximum potential sunshine duration (h/day):here 13 hShademin: Minimum potential sunshine duration (h/day):here 0 hShade: Map stack of potential sunshine durations (h/day)

The calibration parameterRexp (–) is used to define therange of the temporal and spatial variableRexpMap which iscalculated for each day and each grid cell.

Following Braun et al. (1993) the dimensionless param-eter Rice (>1, –) is introduced to the temperature-indexmethod for the simulation of ice melt of the snow-free partsof glaciers. Melt conditions of ice under debris-cover mustbe considered separately from the conditions of debris-freeglaciers. Depending on the thickness of the debris cover,melting can be decoupled from the current meteorologicalsituation. Rana et al. (1996) reported an average thicknessof debris cover of 0.5–1.0 m for the ablation area of Lirungglacier and Langtang glacier (Fig. 1) and therefore a reduc-tion of melt is assumed (see Popovnin and Rozova, 2002).The additional parameterRdebris (<1, –0) in Eq. (4) wasintroduced following Braun et al. (1993).Rdebris andRiceare set 1.0 if snow melt is calculated or melt of debris-freeglaciers is simulated (Rdebris=1.0,Rice>1.0). Melting of firn(“old snow” stored in the snowpack) is calculated the sameway as melting of the seasonal snowpack. The model doesnot account for the movement of ice.

For each cell with a glacier cover, the mass balance is cal-culated based on the simulated snow accumulation and thesimulated snow melt and ice melt. The annual mass balanceof the entire glacier is the arithmetic mean of the mass bal-ances of all cells belonging to the same glacier.

4.3 Soil routine

Typical soils with clearly identifiable horizons are rare in thecatchment, thus, the main function of the soil routine is tostore near-surface water and allow evaporation before waterreaches the runoff generation routine. The output of the snowroutine is the input into the soil routine, which was adoptedfrom the HBV model (Bergstrom, 1976, 1992). This routinewas not changed compared to previous application of TACD.

4.4 Runoff generation routine and runoff routing

The runoff generation routine is the core of the model. Itsstructure has to be modified for each catchment, dependingon process knowledge and data availability (Uhlenbrook etal., 2004). The routine uses as patial delineation of units withthe same dominating runoff generation processes. Usually,the units have to be delineated with supporting informationon runoff sources or flow paths. These insights into catch-ment characteristics are gained from tracer experiments andhydrochemical data. In the case of Himalayan catchmentsthese data are not available. However, as shown in Braunet al. (1993), a simpler lumped runoff generation routine isnot able to simulate discharge without statistical corrections.With the distributed approach we can implement distinct con-ceptualised runoff generation processes of obviously differ-ent units. Since tracer data are missing only basic units can



be distinguished based on topography and physiographic in-formation.

Units with similar dominating runoff generation behav-ior were delineated using maps of topography and land use,aerial photographs and a DEM as well as experiences gainedduring field visits. The delineation is the basis for translatingthe dominating runoff generation processes into the runoffgeneration routine of the model, as it defines the model struc-ture and parameterization. The way the delineation wascarried was fed also by extensive experiences obtained inhydro-glaciological research projects at European and tropi-cal glaciers in the last decades. The plausibility of the spa-tial delineation of units was discussed with local experts sup-ported by field observations during different seasons. Fourunits were distinguished (Fig. 2):

1. non-glacier area (nRGType 1, 53% of catchment area),

2. glacier area (nRGType 2, 45%),

3. glacier area with an inclination of less than 3◦ and debriscover (nRGType 3, 1%), and,

4. valley bottoms with an inclination of less than 8◦

(nRGType 4, 1%).

Sequentially connected or overflowing reservoirs simulatethe runoff processes of unit types 1 and 2. In these unitsupper and lower storages exist which are vertically linkedvia a constant percolation rate. For conceptualization of therunoff generation processes of the third unit type (nRGType3), only a single storage is used. Storage capacity is lim-ited by an upper limit. Runoff of this storage is computedby applying a storage coefficient with additional water if thestorage content exceeds the storage capacity. This concep-tualization is based on the assumption that the large valleyglaciers can store a great amount of water in pools or smallsub- and supra-glacial lakes (Jansson et al., 2002). Thus, theupper limit of the storage capacity is larger than the upperlimits of the storages of unit types 1, 2 or 4.

A fourth unit (nRGType 4) was identified based on infor-mation acquired during field visits and on aerial photographs.At the valley bottom with an inclination of less than 8◦ it isconsidered to exist an aquifer made of glacial moraine andgravel beds where water can be stored. The same structureof storage as in nRGType 3 is used to simulate the hydrologi-cal processes but with different parameterization for differentflow dynamics.

Once water fluxes have reached stream cells, the generatedrunoff is routed to the outlet of the catchment during the sametime step. Since the time step is one day this can be justifiedin alpine environments where river flows are turbulent andfast.

5 Model application

The model was tested in the validation period 1987–1993with the best parameter set obtained in the calibration period1993–1997 by manual trial and error technique. The later pe-riod was chosen for model calibration because the quality ofmeteorological data is better during this period. In addition tothe visual inspection of the simulated hydrographs four dif-ferent objective evaluation criteria were used: The model ef-ficiencyReff (Nash and Sutcliffe, 1970), the model efficiencyusing logarithmic discharge values logReff, the coefficient ofdeterminationR2, and the volume errorVE, which is the ac-cumulated difference between simulated and measured dis-charge of a defined period. The logReff is particularly usefulto evaluate low flows as it emphasizes on low discharges.

The initial model parameter set was estimated accordingto basin characteristics, available data from literature (e.g.snow routine) or was derived based on experiences from pre-vious TACD-applications to other basins. In particular, theparameter values of the soil routine and runoff generationroutine were defined such that our understanding of the dom-inant process behavior is represented. Thus, for instance theoutflow parameters of the reservoirs of the runoff generationroutine (cf. Fig. 2) were estimated that the simulated runoffdynamics agree with our process knowledge. To get reason-able initial storage volumes the years before the calibrationperiod were modeled using meteorological data of the vali-dation period. Calibration runs were performed by adjustingonly parameters of the snow and glacier routine and the pre-cipitation input parametersSFCF andPCF. The initial pa-rameters of the soil routine and the runoff generation routinewere not further adjusted as they were defined to representour process understanding.

Glaciers are sources of runoff water, but their contributionis hard to estimate from discharge and scarce meteorologicaldata alone. An overestimation of melt water production can,for instance, compensate for underestimated basin precipita-tion, or vice versa. Therefore, good simulation results canoccur for the wrong reasons. In order to improve the simula-tions, data about the glacier mass balances in addition to dis-charge were considered in the calibration procedure to findthe best parameter set (Table 4) of the snow and glacier rou-tine. A parameter fine-tuning or investigations of the modeluncertainties (equifinality problem, GLUE or other method-olgies) were beyond the scope of this study. It is worth notingin this respect that the simulation of e.g. the calibration pe-riod (5 years) took about 30 min (desktop PC, CPU 1.6 Ghz),because of the integration of the model into the GIS whichmade the model simulation slow.

6 Simulation results

The model results show that about 47% of the basin pre-cipitation falls as snow and 53% as rain, even though 74%



Table 4. Optimized parameter set of TACD for the Langtang Khola catchment.

Parameter Description Determination Value Unit

Precipitation correction and regionalization

PCF Precipitation correction factor for rain Calibration 1.05 (–)PGrad Vertical precipitation gradient Calibration (Externally defined, based on literature) 0.04 (% 100 m−1 100−1)

PHorizGrad Horizontal precipitation gradient Calibration (Externally defined, based on literature) −0.03 (% 1000 m−1 100−1)

SFCF Snowfall correction factor Calibration 1.2 (–)

Temperature regionalization

TGrad Vertical temperature gradient Calibration (Externally defined, based on literature)−0.5 (◦C 100 m−1)

Potential evaporation calculation and regionalization

ETmax Maximum of potential evapotranspiration Calibration (Externally defined, based on literature) 2.2 (mm d−1)

ETGrad Vertical evapotranspiration gradient Calibration (Externally defined, based on literature)−0.01 (% 100 m−1 100−1)

Snow and glacier routine

TT Threshold value of temperature for snowfall also gen-eral temperature correction

Calibration (Externally defined, based on literature) −0.2 (◦C)

CFMAX Degree-day factor Calibration 7.0 (mm◦C−1 d−1)

CWH Water holding capacity of snow Literature (Bergstrom, 1992) 0.1 (–)CFR Coefficient of refreezing Literature (Bergstrom, 1992) 0.05 (–)Rexp Multiplicative factor for cells with maximum potential

sunshine durationCalibration 1.3 (–)

Rice Multiplicative factor to account for accelerated meltover ice as compared to snow

Calibration 1.4 (–)

Rdebris Reduction factor of glaciermelt over debris-coveredparts of the glacier

Calibration (Externally defined, based on literature) 0.3 (–)

Soil routine

LP Reduction parameter of field capacity Literature (Menzel 1997) 0.6 (–)

Non-glacier area (nRGType 1)

FC1 Maximum soil moisture storage (field capacity) Calibration 20 (mm)BETA1 Empirical parameter Calibration 2.0 (–)

Glacier area (nRGType 2)


Glacier area with inclination less 3◦ and debris cover (nRGType 3)


Valley bottom with inclination less 8◦ (nRGType 4)


Runoff generation routineNon-glacier area (nRGType 1)

US K1 Storage coefficient of upper storage Calibration 0.13 (d−1)

LS K1 Storage coefficient of lower storage Calibration 0.005 (d−1)

US P1 Percolation capacity Calibration 1 (mm d−1)

US H1 Limit of upper storage Calibration 100 (mm)

Glacier area (nRGType 2)

US K2 Storage coefficient of upper storage Calibration 0.1 (d−1)

LS K2 Storage coefficient of lower storage Calibration 0.02 (d−1)

US P2 Percolation capacity Calibration 3 (mm d−1)

US H2 Limit of upper storage Calibration 200 (mm)

Glacier area with inclination less 3◦ and debris cover (nRGType 3)

GlacierLSK Storage coefficient of glacier storage Calibration 0.01 (d−1)

GlacierLSH Limit of glacier storage Calibration 3000 (mm)

Valley bottom with inclination less 8◦ (nRGType 4)

ValleyLS K Storage coefficient of valley storage Calibration 0.01 (d−1)

ValleyLS H Limit of valley storage Calibration 1000 (mm)

Routing routine

MaxBas Empirical parameter Set a priori 1 (–)



Table 5. Comparison of evaluation criteria of simulations with TACD and HBV-ETH. Calibration period,1993-1997, and validation period,1987–1993.

TACD HBV-ETH

Reff (–) logReff (–) R2 (–) VE (mm/a) Reff (–) logReff (–) R2 (–) VE (mm/a)

1987/88 0.58 0.05 0.87 97 0.26 −2.83 0.78 2581988/89 0.72 0.32 0.90 40 0.37 −1.45 0.81 1611989/90 0.68 0.45 0.84 50 0.37 −0.49 0.77 201990/91 0.20 0.40 0.88 −30 −0.33 −0.38 0.80 −1361991/92 0.30 0.59 0.82 −9 −0.05 −0.44 0.68 801992/93 0.76 0.68 0.91 37 0.40 −0.60 0.76 1071993/94 0.85 0.75 0.89 57 0.59 0.07 0.75 1231994/95 0.87 0.80 0.88 −8 0.83 0.63 0.91 1261995/96 0.53 0.31 0.72 92 0.29 −0.38 0.62 811996/97 0.46 0.84 0.76 −8 0.18 0.15 0.63 46

of precipitation fall in the relatively warm monsoon period.This demonstrates the important role of snow and ice storageand melt for the water balance in this basin with a glacier-ization of about 46%. Basin precipitation (432 to 668 mm/a)and discharge (470 to 694 mm/a) are by far the largest com-ponents, followed by glacier melt (274 to 422 mm/a). Thechange in storage (135 to 279 mm/a) summarizes water stor-age in groundwater, snow pack or ice and soil. Actual evap-otranspiration accounts for 12 to 17% of the input precipita-tion. The first two hydrological years of the calibration pe-riod show a good agreement of the simulated and measureddischarge (Table 5). The dynamics of the hydrograph arereproduced satisfactorily using the modified TACD. Fromthe hydrological year 1995/96 onwards the model overesti-mates discharge. Figure 3 shows comparisons of simulatedand measured discharge data.

The winter discharge is simulated well in the years1993/94, 1994/95 and 1996/97 with logarithmic model ef-ficiencies (logReff) from 0.75 to 0.84 (Table 5). However,the simulated onset of discharge in May and June is de-layed in most of the years by some 10 days and, conse-quently, discharge is underestimated at the beginning of themelt/monsoon period. Additionally, discharge is generallyoverestimated in the second half of the melt/monsoon period.

The glacier mass balances are calculated for the cells inwhich the point measurements took place. Figure 4 showsthat the balances generally are underestimated both in theablation area and in the accumulation area.

However, the simulated balances are in the right order ofmagnitude. Fujita et al. (1998) determined the aerial averagemass balance of Yala glacier for the investigation period ofabout 20 weeks as−357 mm; the simulated mass balance forentire Yala glacier is−332 mm for the same period, whichcompares favorably considering the accuracy of mass bal-ance measurements.

Measured winter discharge is higher in the validation pe-riod, with an average from November to the end of April of3.35 m3/s compared to the average of 2.73 m3/s in the cali-bration period. The modified TACD underestimates winterdischarge in all hydrological years of the validation periodand tends to overestimate discharge in the melt/monsoon pe-riod. These compensating effects are the reason for the rela-tively small volume errors (Table 5).Reff and logReff, how-ever, show a drop in performance during the validation pe-riod compared to the calibration period, whereas the coeffi-cient of determination indicates a strong connection betweenmeasured and simulated hydrograph.

The comparison of measured (−2240 mm at 5250 m a.s.l.and 510 mm at 5560 m a.s.l., on Yala glacier, Braun etal., 1993) and simulated (−1872 mm at 5250 m a.s.l. and486 mm at 5560 m a.s.l.) mass balances of the validation pe-riod is satisfactory and points towards that the parameteriza-tion of the snow and glacier routine of the calibration periodis representative for the entire simulation period.

7 Discussion

7.1 Pre-processing of input data

The second order polynomial function used to bridge gapsin temperature time series can reproduce the measured tem-perature course and even single patterns were modeled real-istically. The simulation of the monsoon season is generallybetter than the simulation of the dry season. The coefficientsof determination (R2) of the years in which the method wasapplied varied between 0.70 and 0.91. Note that the simu-lated values were used only to fill the missing values in thetime series.

The measured precipitation values are compared with thesimulated values of June and July 1997 in Fig. 4 for theillustration of the precipitation simulation results. Some



Fig. 3. Simulation results of TACD and HBV-ETH model.

of the precipitation sums are reproduced well, e.g. around27/7/1997; others are redistributed over several days, e.g.around 29 June 1997. However, there are some significantdeviations between the measured and the simulated precip-itation values. The reconstruction of the precipitation sumsis based on probabilities and it is not an extrapolation likethe temperature extrapolation where reliable correlations be-tween SGHU station and reference stations are a prerequi-site. Nevertheless, it is concluded that the extrapolated timeseries are able to describe the actual situation adequately andthe main patterns, e.g. timing and magnitude, of the mea-sured precipitation sums are reproduced well by the algo-rithm. The simulated daily precipitation sums were adjustedto the actual precipitation sums using the ratio of calculatedand measured monthly precipitation sums and therefore thedeviations of the monthly amounts of precipitation are notsignificant. If data availability is not sufficient, monthly pre-cipitation sums had to be estimated from other years. The

temporal shift of the simulated precipitation sums does notnecessarily influence the discharge simulation. If the storagecapacities of snow pack or of the storages of the runoff gen-eration routine are not exceeded, additional water is added tothese storages and contributes to runoff generation with someretardation. Thus, errors in temporal assignment of simulateddaily precipitation values can be tolerated by the model toosome extent. It must be further considered that precipitationevents in the same region do not always occur at the sametime or are stationary throughout the event. If precipitationoccurs at a reference station at one day the algorithm im-plies that there is rain- or snowfall at the SGHU station onthe same day, whereas the precipitation event might have oc-curred in reality on the following day at the SGHU station.This causes temporal inaccuracies of the simulated precipi-tation sums even if there is a strong connection of the meteo-rological conditions between the region and the target station(SGHU station).A comparison of the simulated hydrographs



5131 m / 5150 m

5183 m / 5190 m

5257 m / 5230 m

5296 m / 5280 m

5336 m / 5350 m

5382 m / 5370 m

5423 m / 5390 m

Areal average

Gla

cier

mas

s ba

lanc

e (m

m)

-2500

-2000

-1500

-1000

-500

0

500

1000

1500Measured glacier mass balanceSimulated glacier mass balance

Fig. 4. Comparison of simulated glacier mass balances with theTACD model and measured glacier mass balances at different alti-tudes on Yala glacier from 19/05/1996 to 6/10/1996 (Fig. 1). Theblack bars are behind the gray bars and both start at 0.00 mm. Firstnumbers of x-axis are the altitudes of the 200×200 m2 cells inwhich the balance was measured. Second numbers show the alti-tudes of the observation points.

calculated based on measured daily precipitation sums andon the synthetic daily precipitation sums in Fig. 5 shows thatthe differences are small for most of the times.

The hydrological year 1995/96 was chosen for that com-parison because this year shows neither missing values noraccumulated precipitation sums over several days in the orig-inal measured time series. Thus, no pre-processing was ap-plied for this year and therefore it is applicable for the eval-uation of the precipitation pre-processing method. The mea-sured precipitation event of 25/7/1996 amounts to 25 mmwhile the method to fill the missing values distributes thisamount over previous days as the measured precipitationsums of the reference stations implies. The discharge peakcalculated with the measured precipitation sums is thereforenot found in the simulated discharge with synthetic precipi-tation input. To summarize, the pre-processing methods fortemperature and precipitation data are generally appropriatefor filling missing values in the time series and for redis-tributing precipitation sums. They can further be used forreliability assessment of measured data.

7.2 Regionalization of input data and evapotranspiration

The regionalization of air temperature and precipitation inthe model itself with fixed vertical and horizontal gradientsseems more problematic because seasonal variations of thegradients are not considered. As precipitation and air temper-ature measurements are available for only one station in thecatchment the use of sophisticated regionalization methods(e.g. Kriging or Inverse Distance Weighting) for the meteo-rological input data are prevented. The introduction of the

02/06/97 16/06/97 30/06/97 14/07/97 28/07/97

Pre

cipi

tatio

n (m

m/d

)

0

5

10

15

20

25

30simulated precipitationmeasured precipitation

Fig. 5. Comparison of measured daily precipitation sums (red bars)and simulated daily precipitation sums (green bars).

horizontal gradient that is substantiated by the topographicsituation (Shiraiwa et al., 1992), however, enables a more re-alistic distribution of the basin precipitation with less precipi-tation in the north of the catchment. There is only one stationin the Langtang Khola catchment itself (located in the mainvalley close to the outlet!) which is meant to represent the en-tire meteorological situation of the catchment. The absenceof further data makes the evaluation of the representativity ofthis station hard and an independent evaluation of the spatialand temporal extrapolation of the precipitation data impossi-ble.

For the calculation of the potential evapotranspirationa simple sinusoidal approach was chosen (Hottelet et al.,1993). Only 14.6% of the entire catchment area is below4500 m and as evaporation generally decreases with eleva-tion and with the presence of snow and ice covers, this termplays a minor role in the water balance (Lang, 1981). Further,no additional data are required to simulate evapotranspirationfollowing this approach.

7.3 Temporally and spatially distributed modeling of snowand ice melt

Since there is no basis for a direct evaluation of the dis-tributed melt modeling approach we demonstrate how thealgorithm works in general. Figure 7 shows the spatial dis-tribution of RexpMap for the winter and the summer solstice.The spatial distribution ofRexpMap shows realistic patterns:on 21 December the lowest values ofRexpMap can be foundat the north-oriented slopes at the valley bottom, whereasthe highest values occur at the peaks and at highly elevatedsouth-oriented plateaus. The zenith angle of the sun is closeto 0◦ in the Langtang valley at the summer solstice due toits location close to the Tropic of Cancer (23.5◦ N). Thus,the map of 21 June shows the highest values at cells withnorthern and southern orientation and at the peaks or ridges.



01/10/95 01/12/95 01/02/96 01/04/96 01/06/96 01/08/96 01/10/96

Dis

char

ge (

m3/

s)

0

5

10

15

20

25

30

35

Measured dischargeSimulated discharge with measured precipitationSimulated discharge with extrapolated precipitation

Pre

cipi

tatio

n (m

m)

0

5

10

15

20

25

30

35

Measured precipitationExtrapolated precipitation

Fig. 6. Comparison of the influences of different precipitation inputs on the discharge simulation. The hydrological year 1995/96 was chosenbecause of the data availability.

Fig. 7. Spatial distribution of the correction factor for the degree-day methodRexpMap(–) at the winter solstice (21 December, left) and thesummer solstice (21 June, right).

The lowest values are at east- or west-oriented slopes. Tosummarize, the introduction ofRexpMap enabled spatial andtemporal distributed modeling of snow- and ice melt basedon topographic and astronomic information.

The estimated degree-day factor (product ofCFMAX,RexpMap, Rice andRdebris) is about twice as high as in degree-day model applications in catchments in the European Alps(e.g. Braun and Aellen, 1990). Observations of the sur-face glacier melt rates are reported for the Yala glacier at5100 m a.s.l. (Fig. 1) by Motoyama and Yamada (1989) forthe period 23 August–3 September 1987, and a value of12.7 mm◦C−1 d−1 (based on hourly values) was observed.Here, our simulated value (debris free glacier,Rdebris=1) forthe same area amounts to 11.9 mm◦C−1 d−1 and compares

well with the observations. Further literature values of sur-face glacier melt rates on Khumbu glacier (Mount Everestregion) were even larger with 16.9 mm◦C−1 d−1 for bare ice(Kayastha et al., 2000).

About 19% of the glaciers in the Langtang Khola catch-ment are covered by debris, most of which lie in the low-est parts of the basin. The parameterRmultd was introducedto account for the reduction of ice melt beneath the debrislayer. The sensitivity analysis (Fig. 7) of this parametershows that discharge is generally underestimated if melt issuppressed totally (Rmultd=0.0) and overestimated if debris-covered glaciers are treated the same way as clean glaciers(Rmultd=1.0).



In the sample year 1994/95 volume losses of 100 mm/a oc-cur when settingRmultd=0.0, discharge is overestimated byabout 269 mm/a ifRmultd=1.0.Rmultd=0.3 was chosen basedon the experiences of Popovnin et al. (2002) at Djankuatglacier, Caucasus. They found a reduction of melt under de-bris layers of 50 to 70 cm of approximately 70% comparedto the melt rate of bare ice.

7.4 Goodness of simulation of discharge and glacier massbalances

The overall performance of the model for the hydrologi-cal years 1993/94 and 1994/95 can be considered as good(e.g. Rango, 1992). However, there are various reasons forpartly insufficient simulation results (i.e. simulation of onsetof monsoon period, simulation of single peaks, and simula-tion of glacial mass balances) that will be discussed in thefollowing.

First, one reason for the differences of the simulatedand measured glacier mass balances is the cell size of200×200 m2 used in the model. That means that point mea-surements are compared with simulated mass balances of anarea of 40 000 m2. The altitudes of the cells and of the pointmeasurements differ, because the altitude of the grid cells isan average value of the 40 000 m2 area. Therefore, differ-ent meteorological conditions prevail at the cells and at theobservation points.

Second, the spatial resolution of 200×200 m2 and the rel-atively large daily time step causes a damping of the fastrunoff components. A reduction of the time step from dailyto hourly intervals was not possible because measurementswere only available in daily resolution. A coarser spatialresolution would improve the simulation of fast runoff com-ponents, because traveling distance is larger per time step.This is, however, problematic for the simulation of snow- andice melt because detailed physiographic information gets lostwith a coarser spatial resolution.

Third, the simulation of the onset of discharge at the be-ginning of the monsoon period is delayed by some ten days.Redistribution of snow via possible avalanches and wind driftmust also have a significant impact on the melting conditionsof the early monsoon period. If snow, which has fallen duringwinter period is redistributed into lower altitudes it will startmelting earlier causing higher runoff generation in the be-ginning of the monsoon season. However, mass movementsthrough avalanches and wind drift could not be considered inthe model. In the future, remote sensing data might help toestimate the impact of such processes on the aerial redistri-bution of snow and, consequently, lead to more realistic inputdata sets.

The additional glacier mass balance data of the calibrationand validation period enabled a more reliable adjustment ofthe parameter of the snow and glacier routine. Thus, the massbalance data were taken as valuable additional criteria besidedischarge data for calibration and validation of the model.

01/10/94 01/12/94 01/02/95 01/04/95 01/06/95 01/08/95 01/10/95

Dis

char

ge (

m3/

s)

0

5

10

15

20

25

30

35Measured dischargeSimulated discharge with Rmultd = 0.3Simulated discharge with Rmultd = 0.0Simulated discharge with Rmultd = 1.0

Fig. 8. Sensitivity of the parameterRmultd, Langtang Khola catch-ment.

7.5 Analysis on dominant runoff components

The intra-annual distribution of water with the spatially dis-tributed reservoir concepts of the model enabled good sim-ulation results as expressed by the objective evaluation cri-teria. The composition of runoff shows seasonal variations(Fig. 8).

The high flow season is dominated by the outflow fromthe upper storages. The low flow is mainly a superpositionof runoff of the lower storages of nRGType 1 and 2 and ofthe storages of nRGTypes 3 and 4. Outflow from the stor-ages of nRGTypes 3 and 4 are important components for themaintenance of winter discharge although the area of theseHRUs is small (2%) compared to the contribution areas ofnRGType 1 or 2. Water coming from units 1 or 2 has to passunits 3 or 4 on its lateral flow and thus fills these storagesbeside precipitation and snow melt.

A conceptual storage approach that is not too complex butrepresents the main runoff generation processes of the Lang-tang Khola catchment can be considered as suitable. Withthese sequentially linked storages it was possible to simulatewater storage in glaciers and the underlying hard rock withdifferent hydraulic behavior. Thus, it was not necessary toapply a synthetic constant value for base flow as in Braun etal. (1993). The topographic information combined with landuse and surface characteristics for the delineation of hydro-logical response units enabled a more realistic runoff gener-ation simulation in the remote study area, where additionalinformation (e.g. tracer experiments) were not available.

7.6 Comparison of TACD with the HBV-ETH model

The simulation results of TACD are compared with the re-sults of the HBV-ETH model to provide a broader context forassessing TACD. The HBV-ETH precipitation-runoff modelis a version of the widely used HBV model (Bergstrom,1976, 1992) ith special features to calculate the runoff of



Pro

port

ion

(%)

0

20

40

60

80

100

01/10/93 01/12/93 01/02/94 01/04/94 01/06/94 01/08/94 01/10/94

Inflo

w in

to r

iver

net

wor

k (m

m/d

)

0

10000

20000

30000

40000

50000

Outflow lower storage nRGType 2

Outflow lower storage nRGType 1

Outflow upper storage nRGType 2

Outflow upper storage nRGType 1

Outflow storage nRGType 4

Simulated discharge

Measured discharge

Outflow storage nRGType 3

Fig. 9. Contribution of each runoff component to the entire runoff in percent of the entire runoff (upper graphic) and as absolute values(lower graphic) for the hydrological year 1993/94, Langtang Khola catchment.

glacierized catchments (Braun and Renner, 1992; Hottelet etal., 1993) It has been applied successfully in various Alpineand Himalayan catchments (Braun et al., 2000). Altitudebelts and four orientation classes (North, South, East, West)are used for semi-distributed computation in the snow andglacier routine. The glacier area is considered as a percent-age of the area of each altitude belt and orientation class. Atemperature-index method is used for the calculation of snowand ice melt as in TACD, whereas the computation of melt ofdebris-covered glaciers is treated the same way as of debris-free glaciers. Therefore, HBV-ETH does not distinguish thedifferent glacier surface conditions for melt water simulationas TACD does. The soil routine and the runoff generationroutine are not spatially distributed, thus treated as lumped.An upper (two outlets) and a lower (one outlet) storage forthe entire catchment simulate the runoff generation. For themodel comparison the same calibration and validation datasets were used. The calibration efforts for the HBV-ETHwere at least similar to the ones for the TACD model.

The comparison of the evaluation criteria (Table 5) showsthat TACD generally performs better. Figure 10 shows thatthe monsoon season in July and August 1994 is simulatedreasonably well by both models. However, the HBV-ETHmodel underestimates the discharge both at the beginning of

the monsoon season and during the post-monsoon season.These observations are typical for all simulated hydrologicalyears.

It is difficult to compare the models due to their differ-ent spatial discretization. Therefore, Fig. 10 shows the out-put from the snow and glacier routine and the input into therunoff generation routine of TACD as an average value of allHRUs and the storage levels related to the entire catchmentfor summer 1994. For the output from the snow and glacierroutine HBV-ETH delivers higher values if precipitation con-tributes significantly to the composition of the output of thesnow and glacier routine because there is no negative hori-zontal gradient for the calculation of basin precipitation asin TACD (see Meteorological input data). Therefore, basinprecipitation calculated by HBV-ETH is larger than the onecalculated by TACD. At the beginning and at the end of themonsoon season more water is retained in the soil routine inthe HBV-ETH model than in the TACD model. The inputinto the runoff generation routine does not contribute to thefilling of the upper storage of the HBV-ETH model until themiddle of June 1994. The water is directed to the lower stor-age because the input into the upper storage is smaller thanthe percolation rate (CPERC) of 1.5 mm d−1. Therefore, theonset of discharge at the beginning of the monsoon season



Fig. 10. Comparison of the modeling results of the HBV-ETH model and the TACD model for summer 1994.

simulated by the HBV-ETH model is delayed more than thesimulated discharge of the TACD model. The upper storageof the HBV-ETH model is discharging continuously from themiddle of August until the end of September 1994, whereasthe upper storages of the TACD model also contribute butremain more or less at the same level. This causes the fastdecline of discharge simulated by the HBV-ETH model atthe end of the monsoon season. The lower storage outflowof HBV-ETH maintains the winter discharge and the storageis recharged during the monsoon season. During the calibra-tion of the HBV-ETH model it was realised that the modelcan either be calibrated to simulate the high flow season wellor the low flow season in winter, but it was not possible toget a good simulation for the entire hydrological year. Themore sophisticated and distributed runoff generation routine

of TACD enables the storage of water for maintaining thehigh winter discharge and at the same time simulating thedischarge conditions of the monsoon season reasonably well.

To summarize, it can be stated that both models are appli-cable to the Langtang Khola catchment with success consid-ering the assumptions and quality of the input. Both modelsare able to simulate the melting of ice and snow appropri-ately. The advantages of the TACD model are the better redis-tribution of the water stored during monsoon season whichresulted a better simulation of the onset of discharge in latespring and in better simulations of recession and low flowperiods during the winter (Figs. 3 and 10).



8 Concluding remarks

The study demonstrates a suitable way to simulate dailydischarge values by incorporating basic process knowledgefrom remote Himalayan headwaters with limited data avail-ability. The distributed, conceptual hydrological modi-fied TACD model has proven to be useful in such a data-scarce, high-alpine region. The combination of statisticaldata pre-processing methods for temperature and precipi-tation with simple, robust modeling approaches enables awidely process-based, distributed simulation of the waterbalance what is a clear improvement to previous modelingefforts. In comparison to previous applications of conceptualprecipitation–runoff models to the Langtang-Khola catch-ment, we are now able to simulate the entire annual hydro-graph without empirical adjustments like the application ofconstant base flow values. The more process-based char-acter of the newly introduced runoff generation and glacierroutines enables the redistribution of water which is storedduring monsoon season to simulate the relatively high winterdischarge. The model uses topographical and physiograph-ical information for the simulation of discharge and runoffgeneration processes to compensate for the lack of measure-ments, e.g. the use of calculated potential sunshine durationto compensate for the lack of further climatic stations in thecatchment.

The analysis of the observed climate and discharge datatogether with the model application sharpened our questionsfor a number of needed process research studies in this en-vironment: (i) The origin of the constant winter base flowneeds to be better understood. In particular the role ofgroundwater versus melt water stored in the intra- and sub-glacial systems should be investigated using combined tracerand geophysical techniques. (ii) The role of redistribution ofsnow through and wind drift and avalanches that results inmore snow in the lower valley need to be quantified. A com-bined remote sensing and ground mapping approach seemsto be promising to observe the space and time variability ofthe snow cover. (iii) Additional, the role of frozen soil withregards to runoff generation processes (high flows and lowflows) is poorly understoond. Also suitable ways to translatethe related processes into the model space need to be devel-oped.

Acknowledgements.The authors are grateful to SGHU staff,who collected and prepared the data. Further data was pro-vided by ICIMOD, Nepal. The help of L. Wissmeier (ETHLausanne, Switzerland) during model development is gratefullyacknowledged. Finally, the authors wish to thank S. Maskey(UNESCO-IHE, Delft, The Netherlands) for critical comments ona previous version of the manuscript. The authors are grateful tothe German IHP/HWRP National Committee for their financialsupport of the research work.

Edited by: A. Montanari

References

Alpenvereinskartographie: Alpenvereinskarte, Langthang Himal-Ost, No. 0/11, Austrian Alpine Club, 1990.

Bergstrom, S.: Development and application of a conceptual runoffmodel for Scandinavian catchments, Bull. Series A52, Universityof Lund, 134 pp., 1976.

Bergstrom, S.: The HBV-model – its structure and applications,SMHI Reports RH No. 4, Norrkoping, Sweden, 1992.

Braun, L. N. and Aellen, M.: Modeling discharge of glacierizedbasins assisted by direct measurements of glacier mass balance,IAHS Publ. No. 193, 99–106, 1990.

Braun, L. N., Grabs, W., and Rana, B.: Application of a conceptualprecipitation runoff model in the Langtang Khola basin, NepalHimalaya, IAHS Publ. No. 218, 221–237, 1993.

Braun, L. N., Hottelet, Ch., Weber, M., and Grabs, W.: Measure-ment and simulation of runoff from Nepalese head watersheds,IAHS Publ. No. 248, 9–18, 1998.

Dam, J. C. van: Field-scale water flow and solute transport: SWAPmodel concepts, parameter estimation and case studies, Ph.D.Thesis, Wageningen Institute for Environment and Climate Re-search, Wageningen University, The Netherlands, 2000.

Fujita, K., Takeuchi, N., and Seko, K.: Glaciological observationsof Yala Glacier in the Langtang Valley, Nepal Himalayas, 1994and 1996, Bull. Glacier Res., 16, 75–81, 1998.

Grabs, W. and Pokhrel, A. P.: Establishment of a measuring ser-vice for snow and glacier hydrology in Nepal – conceptual andoperational aspects, IAHS Publication No. 218, 3–16, 1993.

Heuberger, H., Masch, L., Preuss, E., and Schrocker, A.: Quater-nary landslides and rock fusion in Central Nepal and in the Ty-rolean Alps, Mountain Research and Development, 4, 345–362,1984.

Hock, R.: A distributed temperature-index ice- and snowmelt modelincluding potential direct solar radiation, J. Glaciol, 45(149),101–111, 1999.

Hock, R: Temperature-index melt modelling in mountain areas, J.Hydrol., 282, 104–115, 2003.

Hottelet, Ch., Braun, L. N., Leibundgut, Ch., and Rieg, A.: Simula-tion of Snowpack and Discharge in an Alpine Karst Basin, IAHSPubl. No. 218, 249–260, 1993.

Jansson, P., Hock, R., and Schneider, T.: The concept of glacierstorage: a review, J. Hydrol., 282, 116–129, 2002.

Karssenberg, D., Burrough, P. A., Sluiter, R., and de Jong, K.: ThePC Raster software and course materials for teaching numericalmodelling in the environmental sciences, Transactions in GIS,5(2), 99–110, 2001.

Kattelmann, R.: Role of snowmelt in generating streamflow duringspring in east Nepal, IAHS Publ. No. 218, 103–111, 1993.

Kayastha, R. B., Takeuchi, Y., Nakawo, M., and Ageta, Y.: Practi-cal prediction of ice melting beneath various thickness of debriscover on Khumbu Glacier, Nepal, using a positive degree-dayfactor, IAHS Publ. No. 264, 71–81, 2000.

Kayastha, R. B.: Study of glacier ablation in the Nepalese Hi-malayas by the energy balance model and positive degree-daymethod, Ph.D. Thesis, Graduate School of Science, Nagoya Uni-versity, 95 pp., 2001.

Klok, E. J., Jasper, K., Roelofsma, K. P., Gurtz, J., and Badoux, A.:Distributed hydrological modeling of a heavily glaciated Alpineriver basin, Hydrological Sciences Journal – Journal Des Sci-ences Hydrologiques, 46, 553–570, 2001.



Konz, M., Braun, L., Uhlenbrook, S., Shrestha, A.: Runoff fromNepalese Head Watersheds based on Measurements and Mod-elling, IHP/HWRP-Bericht, Heft 4, ISSN 1614-1180, 2006a.

Konz, M., Braun, L., Uhlenbrook, S., Shrestha, A., and Demuth,S.: Regionalisation of a distributed catchment model for highlyglacierized Nepalese headwatersheds, Proc. 5th FRIEND WorldConference, Water resource variability: Processes, analyses andimpacts, 27 November–1 December 2006, Havana, Cuba, IAHSPubl. 308, 454–465, 2006b.

Lang, H.: Is evaporation an important component in high alpinehydrology?, Nordic Hydrology, 12(4/5), 217–224, 1981.

Motoyama, H. and Yamada, T.: Hydrological observations in theLangtang Valley, Nepal Himalayas during 1987 monsoon – post-monsoon season, Bull. Glacier Res., 7, 195–201, 1989.

Nash, J. E. and Sutcliffe, J. V.: River flow forecasting through con-ceptual models, Part I – a discussion of principles, J. Hydrol., 10,282–290, 1970.

O’Callaghan, J. F. and Mark, D. M.: The extraction of drainagenetworks from digital elevation data, Comput. Vision GraphicsImage Process, 28, 328—344, 1984.

Ono, Y.: Glacial fluctuations in the Langtang Valley, Nepal Hi-malaya, Gottinger Geographische Abhandlungen, 81, 31–38,1986.

Ott, B. and Uhlenbrook, S.: Quantifying the impact of land usechanges at the event and seasonal time scale using a process-oriented catchment model, Hydrol. Earth Syst. Sci., 8(1), 62–78,2004.

Popovnin, V. V. and Rozova, A. V.: Influence of sub-debris thawingon ablation and runoff of the Djankuat glacier in the Caucasus,Nord. Hydrol., 33(1), 75–94, 2002.

Ramage, C. S.: Monsoon meteorology, Academic Press, 1971.Rana, B., Nakawo, M., Fukushima, Y., and Ageta, Y.: Runoff mod-

elling of a river basin with a debris-covered glacier in LangtangValley, Nepal Himalaya, Bull. Glacier Res., 14, 1–6, 1996.

Rango, A.: Worldwide testing of the snowmelt runoff model withapplication for predicting the effects of climate change, Nord.Hydrol., 23, 155–172, 1992.

Sakai, A., Fujita, K., and Kubota, J.: Evaporation and percolationeffect on melting at debris-covered Lirung Glacier, Nepal Hi-malayas, 1996, Bull. Glacier Res., 21, 9–15, 2004.

Seko, K.: Seasonal variation of altitudinal dependence of precipita-tion in Langtang Valley, Nepal Himalayas, Bull. Glacier Res., 5,41–47, 1987.

Shiraiwa, T., Ueno, K., and Yamada, T.: Distribution of mass in-put on glaciers in the Langtang Valley, Nepal Himalayas, Bull.Glacier Res., 10, 21–30, 1992.

Tarar, R. N.: Water resources investigation in Pakistan withthe help of Landsat imagery-snow surveys, 1975–1978, IAHSPubl. No. 138, 177–190, 1982.

Uhlenbrook, S. and Leibundgut, Ch.: Process-oriented catchmentmodeling and multiple-response validation, Hydrol. Processes,16, 423–440, 2002.

Uhlenbrook, S., Roser, S., and Tilch, N.: Development of a dis-tributed, but conceptual catchment model to represent hydrologi-cal processes adequately in the meso scale, J. Hydrol., 291, 278–296, 2004.

UNEP/ICIMOD: Inventory of glaciers, glacier lakes and glacierlake outburst floods monitoring and early warning systems in theHindu Kush – Himalayan region [CD-ROM ISBN 92 9115 3591], 2002.

Verbunt, M., Gurtz, J., Jasper, K., Lang, H., Warnerdam, P., andZappa, M.: The hydrological role of snow and glaciers in alpineriver basins and their distributed modeling, J. Hydrol., 282, 36–55, 2003.

Viviroli, D. and Weingartner, R.: The hydrological significance ofmountains: from regional to global scale, Hydrol. Earth Syst.Sci., 8, 1016–1029, 2004,http://www.hydrol-earth-syst-sci.net/8/1016/2004/.

Weber, M.: Aspekte zur Extrapolation von Tagesmittelwerten vonTemperatur und taglichen Niederschlagssummen an hochgelege-nen Gebirgsstationen aus Klimadaten desortlichen Klimamess-netzes in Nepal, internal report Bavarian Academy of Sciences,Commission of Glaciology, Munich, 1997.

Zappa, M., Pos, F., Strasser, U., Warmerdam, P., and Gurtz, J.: Sea-sonal water balance of an alpine catchment as evaluated by dif-ferent methods for spatial distributed snowmelt modeling, Nord.Hydrol., 34(3), 179–202, 2003.

Zingg, Th.: Beziehung zwischen Temperatur und Schmelzwasserund ihre Bedeutung fur Niederschlags- und Abflussfragen, IAHSPubl. No. 32, 266–269, 1951.


http://www.hydrol-earth-syst-sci.net/8/1016/2004/

implementation of a process-based catchment model in a ... · oped for the himalayan conditions. it...

Documents